experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #36386 from ROCmSoftwarePlatform:google-upstream-pr-gpu_unrolling_test

Browse Source 
PiperOrigin-RevId: 292641827
Change-Id: I6e5f6c4030de2a6b7566f7837cc97d8639378fe5
This commit is contained in:
TensorFlower Gardener 2020-01-31 16:48:44 -08:00
commit 0989420102
1 changed files with 24 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
tensorflow/compiler/xla/service/gpu/tests/gpu_unrolling_test.cc
View File

@ -154,12 +154,21 @@ TEST_F(GpuUnrollingTest, DisabledUnrollUnfusedSine) {
auto hlo_module =
ParseAndReturnVerifiedModule(kUnfusedAddModule, config).ValueOrDie();
CompileAndVerifyIr(std::move(hlo_module),
R"(
// Note: On ROCm side, we do bare minimal to make the test pass.
// "sine" function is in different code generation path from nvptx: on
// ROCm platform, it get pulled in from ROCm-Device-Libs, whereas in
// Cuda, generated llvm IR is compiled PTX.
auto expected_ir = is_built_with_rocm_ ? R"(
; CHECK: __ocml_sin_f32
; CHECK-NOT: load float
)"
: R"(
; CHECK: load float
; CHECK-NOT: load float
}
)",
)";
CompileAndVerifyIr(std::move(hlo_module), expected_ir,
/*match_optimized_ir=*/true);
}
@ -179,12 +188,21 @@ TEST_F(GpuUnrollingTest, DisabledUnrollUnfusedCosine) {
auto hlo_module =
ParseAndReturnVerifiedModule(kUnfusedAddModule, config).ValueOrDie();
CompileAndVerifyIr(std::move(hlo_module),
R"(
// Note: On ROCm side, we do bare minimal to make the test pass.
// "cosine" function is in different code generation path from nvptx: on
// ROCm platform, it get pulled in from ROCm-Device-Libs, whereas in
// Cuda, generated llvm IR is compiled PTX.
auto expected_ir = is_built_with_rocm_ ? R"(
; CHECK: __ocml_cos_f32
; CHECK-NOT: load float
)"
: R"(
; CHECK: load float
; CHECK-NOT: load float
}
)",
)";
CompileAndVerifyIr(std::move(hlo_module), expected_ir,
/*match_optimized_ir=*/true);
}